摘要
使用岩石铸体薄片图像对岩石孔隙特征进行分析已经成为国内外石油地质部门常用方法之一,自动精确地分割铸体岩石薄片中的孔隙区域是定量计算孔隙参数的前提。目前传统RGB阈值分割方法精度不高,需要大量人工交互,而一些主流图像分割的深度学习网络泛化性能差,难以运用到实际中。针对这些问题,本文在U-net网络的基础上,提出了一种融合注意力机制和循环残差网络的模型。引入循环残差模块扩展网络深度,又融合了注意力机制模块,增加特征信息的学习权重。采用油田实验室常见的多种铸体薄片进行了实验,均取得了较好的分割结果,验证了本文方法的有效性和泛化性。
It has become one of the common methods for petroleum geology departments at home and abroad to analyze the characteristics of rock pores by using the image of cast rock thin section.Automatic and accurate segmentation of the pore area in cast rock thin section is the premise of quantitative calculation of pore parameters.At present,the traditional RGB threshold segmentation method is not accurate and needs a lot of human interaction.However,the generalization performance of some mainstream image segmentation deep learning networks is poor,which is difficult to apply to practice.To solve these problems,this paper proposes a model based on u-net,which combines attention mechanism and cyclic residual network.The cyclic residual module is introduced to expand the network depth,and the attention mechanism module is integrated to increase the learning weight of feature information.Experiments are carried out on a variety of common cast thin sections in the oilfield laboratory,and good segmentation results are obtained,which verify the effectiveness and generalization of the proposed method.
作者
刘凯文
熊淑华
滕奇志
LIU Kaiwen;XIONG Shuhua;TENG Qizhi(Institute of Image Information,College of Electronics and Information Engineering,Sichuan University,Chengdu 610065,China)
出处
《智能计算机与应用》
2021年第11期68-75,共8页
Intelligent Computer and Applications
基金
国家自然科学基金(62071315)
